…rt docs

Add WithMaxLanes so the Hopper's per-key lane set (and its goroutines and
map entries) is bounded under unbounded key cardinality; keys beyond the
bound fold onto lanes by hash without reordering a key's messages. laneKey
hashes the partition-key string directly, dropping a per-message []byte
allocation.

dispatchBatch now records ErrBatchResultCount on a batch result over-count
so the discarded extra results are visible in traces rather than swallowed.

Correct three inert-surface docs: Deduper is a reserved seam adapted by the
idempotency middleware, not consulted by the Hopper; Batched.SettleBatch is
a one-call settle seam the Hopper does not use (it settles per message); and
remove retry's dead AttemptHeader and WithClock/cfg.now, and make
WithJitterSource's scope honest. Add tests for ReceiveBatch, batch
over-count, WithMaxLanes folding, InProgress/Manual dispositions, and the
cdc DecodeEvent registry-error and RawJSON.As unmarshal-error paths.

Signed-off-by: Joshua Temple <joshua.temple@stablekernel.com>

…ignal

DriveFunc's godoc dropped the broken [Deduper] link for concrete guidance
(use Drive or the idempotency middleware). doc.go now documents the
transactional DriveTx mode alongside the durable and stateless modes.

Drive and DriveTx mark the span with statemachine.exactly_once=false when a
message yields no event id, so the silent loss of dedup is visible in
traces rather than passing unnoticed; EventID's godoc explains it. Add a
DriveTx Save-failure test covering the in-transaction persist-error abort.

Signed-off-by: Joshua Temple <joshua.temple@stablekernel.com>

A transactional inlet is backed by one GroupTransactSession that fences a
single consumer; a second Subscribe could not share it and would silently
return a subscription with no transact session (Begin would then report
not-transactional). Subscribe now rejects the second call with a sentinel
error. Add coverage for the transactional option assembly and the guard.

Signed-off-by: Joshua Temple <joshua.temple@stablekernel.com>

…s): add README and doc.go

jetstream newMessage no longer eagerly builds the header slice; Headers
materializes and caches it on first read, so a message whose headers are
never read pays no per-message header allocation.

Add a README and a doc.go for source/cloudevents, the only source adapter
that lacked both; the package comment moves to doc.go to match the suite's
documentation layout.

Signed-off-by: Joshua Temple <joshua.temple@stablekernel.com>

Signed-off-by: Joshua Temple <joshua.temple@stablekernel.com>